Research On Fruit And Vegetable Recognition Based On Improved Lightweight YOLOv5 Model

Fruits and vegetables play an essential and important role in daily life in China,and the production and sales of fruits and vegetables in China are also increasing year by year.In recent years,as people’s pace of life has become faster and faster,many large supermarkets have equipped self-service checkout machines to adapt to people’s fast-paced consumption.However,in some fruit and vegetable zones,there are still a large number of customers queuing for weighing and settlement.Therefore,consumers hope that supermarkets and farmers’ markets can find ways to change this situation and improve their consumption experience and efficiency when purchasing fruits and vegetables.Although there are also some classification and grading operations in the fruit and vegetable zone,most of them are carried out manually.Although this method is low in cost,it violates the goal of intelligent liberation of productivity,and also has many inevitable drawbacks: there is no unified standard and category determination,which are mostly based on personal experience and instantaneous judgment,and the results are inevitably different from person to person,At the same time,it is also influenced by factors such as the personal emotions and fatigue level of the staff,which inevitably leads to erroneous judgments about the product.There are various types and forms of fruits and vegetables in supermarkets.Mobile tools need to accurately and quickly identify fruits and vegetables,in order to reduce customer queues for settlement and payment of goods,and thus achieve the goal of improving customer consumption experience.This article studies the YOLOv5 object detection algorithm and designs and implements an Android lightweight fruit and vegetable recognition algorithm.The main work of this article is as follows:This article adopts the deep learning YOLOv5 algorithm for research,establishes a fruit and vegetable dataset with packaging bag occlusion conditions,uses the improved model to experiment on the dataset,and analyzes the experimental results.2.Improvements have been made to the YOLOv5 object detection algorithm.In response to the problem of complex computation and large parameter count in the feature extraction network of the original YOLOv5 network,this paper replaces the feature extraction network with a lightweight ghost network,and adds an ECA attention mechanism to the backbone network,reducing the computational complexity of the model and improving the accuracy of model detection.3.Implement Android mobile end deployment of the improved algorithm,and conduct recognition testing on mobile applications.Analyze the test results to prove that the algorithm has good recognition results for common fruit and vegetable targets.